Domain-specific AI knowledge system for the mining industry
MiningGPT began as a public beta on OpenAI’s GPT marketplace, created to test whether mining professionals could get faster access to legal, regulatory and technical information that is often buried in lengthy documents.
The next step was moving beyond the marketplace into a standalone application built around the OpenAI API. This provided greater control over the user experience, authentication, knowledge retrieval and application logic, while validating the broader product architecture.
I am now re-architecting the platform onto AWS Bedrock to create a more robust foundation for domain-specific retrieval and AI delivery. The new architecture uses Bedrock Knowledge Bases with OpenSearch Serverless for vector retrieval, Bedrock Guardrails for output controls, and Terraform-managed AWS infrastructure with least-privilege IAM.
The focus of the re-architecture is not simply changing the underlying model provider. It is creating a system where retrieval, security, infrastructure and application behaviour can be managed deliberately as the platform moves from an experimental AI assistant toward a more production-ready knowledge system.
Architecture evolution OpenAI GPT marketplace beta -> standalone OpenAI application -> AWS Bedrock knowledge platform
Current stack AWS Bedrock | Knowledge Bases | OpenSearch Serverless | Guardrails | Terraform | IAM | Python